A Guide to Writing with AI

Kamal Belmihoub and Constantin Schreiber

Introduction

Students, faculty, staff, and administrators use AI[1] . According to a Lumina Foundation-Gallup 2026 State of Higher Education study, 88% of students use AI, mostly as a partner for learning to check answers for example, while 36% of students use AI to help them write papers (Marken). Early research on the impact of AI on writing and student learning suggests students often find AI useful for invention, drafting, and revision (Cummings; Pritts), but concerns persist over losing one’s agency and voice, and weakening what students learn. The extent to which AI can help you become a better, independent writer remains unclear based on current research, though initial findings indicate that writers can improve aspects of their writing by using AI, such as through improvements in word choice (Cummings et al.; Eslami et al.). We agree with the American Association of Colleges & Universities that we have to remain tethered to values of “compelling teaching, deep learning, human judgment and students’ intellectual independence” and for your writing class, that means prioritizing human inquiry, your original thoughts and perspectives, and your own rhetorical decision-making when writing with AI (“National Survey”).

There are a lot of varied ways instructors incorporate AI into their classrooms. Just take the authors of this chapter: While we both allow, and in specific situations encourage the use of AI in the classroom, we provide different guidelines and expectations around it. For example, Kamal Belmihoub teaches both critical AI literacy and how to write prompts for AI to generate strong outputs. Constantin Schreiber doesn’t do this and mostly focuses on discussing the strengths and limitations of AIs with his students, on raising students’ understanding of AI writing characteristics and ethical use of AI in writing and research, but encourages students to use AI for generating citations, for example. We have different approaches to using it for writing, too. Kamal used an AI to help him outline the writing for this chapter. Constantin does not really use AI (yet?) in his own writing beyond the almost unavoidable automatic spell-checkers. Baruch College offers guidelines for AI use in our institution,[1.1][2.1][3.1] which are flexible enough to accommodate a highly diverse range of approaches to using AI, from approaches that endorse and encourage the use of AI to those who ask you to completely avoid the use of AI inside and outside of the classroom (“AI Use Guidance”).

In this chapter, we provide you with knowledge and tools for developing your own principled approach to writing with AI, including if you decide to avoid it entirely or avoid it only in some situations (Hyland, “Writing”). A potentially helpful way to think about AI is through the lens of “cognitive debt” (Armitage). AI is like using a credit card: you get some immediate benefits at the cost of missing out on the cognitive benefits of using your brain more. As long as you ‘pay’ for this debt on time by using your brain to perform other intellectually-stimulating tasks, you are okay. But if you keep accumulating debt by using AI for everything, the cognitive debt will become overwhelming, to a point where you might not be able to think clearly and make important decisions in your life. So as long as you are able to build cognitive wealth through such activities as close reading, analysis, note taking, summary, paraphrase, synthesis, and so on, some ‘debt’ due to AI use can be beneficial.

Thus, writers should integrate AI into the writing and rhetorical process of text production rather than replacing the writing process. As instructors in the Baruch First-Year Writing Program, we put a lot of emphasis on the process of writing, on all the little steps that are involved in writing a text. We emphasize trying out things, consulting librarians for research, reading, seeking feedback, outlining, making mistakes along the way, struggling, trying again, editing, revising, even starting over. It is our hope that this chapter will provide you with enough (and sufficiently flexible) guidance for developing a comprehensive approach to writing with AI so that you are confident (enough) in your writing, maintain agency, and continue to develop your own voice as a writer.

Before You Start Writing

One element of Baruch’s broader approach to working with AI is that “Generative AI is allowed only where given explicit permission from the instructor. Clearly stated course policy and assignment guidance will be used in determining if the rules of academic integrity have been violated” (“AI Use Guidance”). So, your first step should always be to engage in dialog with your instructor about AI. They should all have AI policies in their syllabus, but these policies may not cover every scenario in detail. Don’t be afraid to talk to your instructors about their approach. They want you to learn and they want you to understand the expectations for their class. We both encourage writing with AI, but we also, like probably all of our colleagues, want you to be in charge and in control of your writing process, we want you to understand and be able to replicate independently the rhetorical moves that you make.

Rather than immediately brainstorming with an AI, we encourage you to first engage in freewriting, expressing your own original ideas before they are influenced by AI. This is a crucial step for maintaining agency over your writing and for bringing in your perspectives and voice. Learning to write with the assistance of AI while making sure that it does not overtake your learning process is a skill that requires attention and practice. It also requires a strong moral compass, especially when you are pressed for time, which we are so often. One way in which you might be able to alleviate the time pressure would be to reach out to your instructor and ask for an extension when you know that you have multiple deadlines and/or tests coming up. Showing your instructor that you are organized and can plan ahead is always a good idea.

Using AI for a Writing Project

Prompt Engineering

Now you are ready to outline and prepare for writing using one or more AIs. When engaging with AI in its current form, writing good prompts is a key to master, especially if you want the AI to help you with the different tasks and challenges throughout the writing process. Designing strong prompts requires carefully considering the following elements: Task, context, role, format, sample, evaluation, and revision. For example, you may design a prompt that tasks an AI with evaluating an outline of a research-based argument paper. After you receive the initial AI output from the prompts below, you can ask follow-up questions using the Socratic Method [2], asking questions to further understanding, to ultimately produce a strong, context-sensitive, well-rounded, and well-reasoned argument. When you prompt an AI chatbot, it is essential to include all components of the prompt-engineering framework to ensure that you obtain meaningful results. If you do not provide sufficient context, odds of inaccuracies and misunderstandings rise. Providing an AI with a text sample that models what you are seeking can also significantly enhance the chances of stronger AI outputs.

Strong AI prompting isn't about typing a quick question and hoping for the best. Writers should give the model sufficient structure to produce high-quality outputs, and the framework gives you that structure. Imagine you ask ChatGPT, "Should I invest in Tesla?” That's a task with nothing else, and the answer you get back will be generic and probably useless for your actual situation. However, adding context (e.g. I'm a 19-year-old college student with $500 in savings, no other investments, no debt, and I'm trying to learn how the stock market works. It’s July 18, 2026.) and role (e.g., “Act as a personal finance educator who explains concepts to beginners without giving specific buy/sell advice”) provides AI with a frame for what kind of response actually fits your life. Format tells the model how to package the answer (e.g., “Give me a 200-word explanation followed by three questions I should ask myself before investing in any single stock”), and the sample component of the framework shows what good looks like. You might write a prompt like this: “Here's an example of the kind of breakdown I find helpful: [insert a website you find delivers helpful explanations]). Without these, you'll get a wall of text or a bulleted list when you may have wanted a conversation, or vice versa.

Evaluation is where you check the output against facts: Does the AI's claim that "Tesla has consistently outperformed the S&P 500" actually hold up when you verify it on a financial site, or is it a confident-sounding hallucination that could cost you real money if you acted on it right before a market correction? Is the current price the AI claims actually current? Finally, revision is the step that perhaps writers might feel tempted to skip. Instead of accepting the first response, you should push back, for example like so: “You didn't address the risk of putting all $500 into one stock and the fact that I need to withdraw the investment in ten years; redo this with diversification in mind”), and the answer gets sharper with each round of revision. You might specifically ask it to draw upon a certain successful hedge fund’s approach or draw upon dominant academic theories among finance researchers. The framework isn't just ideas, it's the difference between using AI as a thinking partner and being misled in all sorts of decisions.

Learning More about a Topic & Evaluating Sources

Before, while, or after outlining your draft you might need to do a bit more research about a topic. You might need to identify sources that can inform you, that can help you ground what you are writing in existing research or opinions. You might want to task AIs to help you with your research plan or with the comparison of sources, for example. One technique that you may use as part of your writing process is to ask AI for possible search terms rather than merely asking for claims and evidence. Another technique might be to ask for narrow researchable questions resulting from current scholarly debates. Asking for sources that matter in such debates, reading those sources yourself, and then asking for comparing perspectives you learned about can advance deeper learning and produce strong and meaningful writing. Take a look at the sample prompts in Table 1 as well.

Table 1:

Prompts for Researching a Topic

AI can help you find sources and get to know about a topic or issue, but it’s important to remember that AI does not think, research, believe, or care in the human sense. It can imitate confidence, coherence, and genre conventions, giving you the illusion that it does a perfect job, but it often includes inaccuracies (Hyland, “EFL”). AI can help writers generate, organize, or refine ideas, but we need to continue to insert human agency and responsibility for truth, evidence, ethics, and judgment into our research and writing. Moreover, AI literacy frameworks emphasize evaluation, context, autonomy, and ethics, all of which are directly relevant to research skills (Allen and Kendeou). The information generated by AI models involves a myriad of choices made by people with a wide range of biases. Repurposing this information by uncritically consuming it and submitting it for others’ (like colleagues, classmates, colleagues, and the writers themselves and their friends and family) consumption may result in perpetuating inaccuracies and inequalities, as well as failures to solve serious problems. AI-generated information cannot be confused with independently verified expert scholarship, industry knowledge, personal lived experiences, and a variety of different forms of knowledge. You should thus still engage in source evaluation, authors’ ethos analysis, and content analysis, you should still consider the age of sources, author’s potential biases, etc.

It is also generally a good idea to triangulate findings – don’t just rely on the sources that AIs provide you with, but to also look for sources using conventional internet search engines as well as browsing the catalog of Baruch’s Newman Library. If all of these three ways of looking for sources produce similar results then you can be fairly certain that you are capturing a lot of the major sources. Often you will find that the findings complement each other, with the sources you find through the library’s search engine often being more scholarly/academic than those a traditional internet search engine might provide you with, and AIs often producing more of a mix of different types of sources. Such triangulation generally produces a stronger list of sources than just relying on one of the pathways for researching information about a topic.

Prompt Engineering Example

In Table 2 below you can find an example of a prompting strategy for outlining and developing ideas for an argumentative research paper. It helps you identify a strong research topic, generate ideas, and advance your intellectual capacity as a learner and thinker throughout the process of reading for and writing your research paper. Before you take a look at these prompts, we want you to note that they are asking the AI to create an outline, not to write full paragraphs or a whole essay. We also want to provide you with a short user manual for them:

First up, it’s important to note that different AIs might produce different outputs in response to the prompts. You might just want to stick to the one you regularly use or the one that you find serves your purposes/writing situation best, but we recommend comparing responses from different AIs that you have access to.

After you read the AI output produced in response to the prompts below, your next steps are to evaluate and revise with new prompts. Reiterate this process of evaluation and revision until you feel like you understand the conversations in the research about the topic (i.e. the intellectual map that captures the key ideas, findings, and authoritative voices on a topic). In other words, you should feel ready to join the conversation. Then, read the actual sources to deepen your understanding. Since many sources will be written by experts for experts you might have to read them strategically, for example by focusing on reading the abstract, the literature review and the results or findings section (see “Reading and Writing” and “Finding and Evaluating Sources” in this book).

We recommend inputting all prompts at once into the AI chatbot, but it is also possible to use them sequentially - doing so would help you see the AI’s revisions in response to each prompt. Pause, however, to stop and evaluate the AI output, whether after each prompt or after you have submitted all the prompts together.

Table 2:

Building an Intellectual Map: A Prompt Library for Outlining

Developing Your Draft with AI: Drafting, Feedback, Revisions, & Editing

When you feel ready to join the conversation, when you have produced an outline with AI, using the above prompts and following evaluation and revision of the outline, you can start writing a draft. As you start writing, you might feel uncertain in your writing decisions. That’s a normal part of the process and we encourage you to engage with that “struggle.” Producing writing isn’t easy for anyone. Nonetheless, you might already want some AI feedback as you put words down. While writing the draft or once you have completed a draft, you may prompt the AI to

●further refine your target audience and purpose, and all of the points raised in the Introduction and Background sections above about rhetorical thinking,

●suggest alternative ways to organize your ideas,

●identify areas of interest to further deepen your knowledge,

●offer critical feedback on any potential logical fallacies and/or

●identify frequent style and grammatical issues that you can address as you revise.

You can find examples for prompts that help you accomplish these goals and more in Tables 3 and 4 below.

Table 3:

Prompt Library for Revising & Editing

A key distinction to maximize AI benefit for writers lies in the difference between revising and editing. Advanced writing requires revision of content, organization, evidence, and rhetorical choices, not just editing word choice, grammar, and spelling. AI can be useful for learning where a claim is weak, where evidence is thin, where a counterargument is missing, or where a paragraph drifts from the thesis. However, it becomes unproductive when you ask only for “make this better” without specifying rhetorical goals: Research on automated writing evaluation has indicated that feedback tools matter only when they are used critically and contextually to rethink and rewrite accordingly (Karatay and Karatay) and writing studies experts warn against reducing writing to surface polish (Hyland, “EFL”). This is why you should follow (or develop yourself) a prompt engineering framework like the one shared earlier along with the central tenets of strong writing discussed above to be able to truly maximize the benefits of AI use. Entry-level positions for many college graduates historically involved collecting, synthesizing, and presenting information to managers, tasks which can now be completed by AI. The extent to which writers will be able to use AI to enhance productivity, make strategic choices, solve complex problems, and stand out as a writer and critical thinker will thus likely largely determine career success moving forward, especially when looking at some research indicating that overreliance on AI is potentially leading to “cognitive surrender” (Shaw and Nave) or “cognitive debt”, which is defined a regular (over)reliance AI replacing “effortful cognitive processes” (like writing) that are needed for critical independent thinking (Kosmyna et al.).

Put succinctly, we suggest you use AI as a smart second reader that helps you notice issues and identify options for addressing them, but never let it replace your judgment, voice, or linguistic identity. Zhang and Hyland show that feedback helps only when writers actively engage with it, so instead of asking AI to “fix” your draft, ask it questions like where readers may be confused, where your evidence feels thin, or where a counterargument is missing. This way, you can respond to the AI feedback and stay in control of your own writing while reflecting on it, learning about it. In addition, Hyland (“EFL”) and Karatay and Karatay suggest that AI improves writing only when writers use feedback critically, so you should ask AI targeted revision questions. To prioritize feedback, we encourage you to leave word- and sentence-level editing until complete revision of all other aspects of the writing process. In other words, your focus as a writer should be on developing your ideas and organizing them well before you begin editing sentences, wording, formatting, and punctuation.

Experts who study the connection between writing and learning have known for decades that the writing process theory emphasizes that meaning is developed through recursive composing; it is not just delivered in finished form (Flower and Hayes). In addition, audience and genre awareness play a significant role in creating drafts with excellent rhetorical judgment, so an AI’s correct language is far from sufficient in ensuring that your writing is meaningful and impactful for you and your readers. Recent scholarship warns that AI often produces a polished but shallow product if writers remain passive (Hyland, “EFL”).

So, it’s essential that writers actively engage with the recursive writing process of ideation, research, drafting, revision, and rhetorical thinking and making rhetorical choices based on audience, genre, purpose, ethos, pathos, logos, exigence, kairos, and stasis judgments. Table 4 provides you with a prompt library for enhancing the rhetorical power of your writing. Note that the AI responses to these prompts should make you engage in the different stages of the writing process and, importantly, they still require you to make decisions, i.e. they help you maintain agency over your writing.

Table 4

Prompt Library for Enhancing the Rhetorical Power of your Writing

Keep in mind that you can work with different AIs and compare their responses!

Conclusion

The term “human-in-the-loop” refers to employees with the capacity to craft strong prompts, generate powerful AI outputs, monitor and evaluate the AI outputs, and be present throughout the process to solve complex problems and make important decisions (Meng). As a student and as a writer, it is critical to position yourself as the human-in-the-loop regardless of your anticipated future job. Writers must learn to use AI in a way that does not replace their own critical thinking or their style and the uniqueness of their voice. Cummings et al. and Jeon et al. express concern that AI can make all writers sound the same and make their voice sound flat. As writing teachers ourselves we have read and have heard reports from our colleagues about “soulless” student writing that relies too heavily on AI. If you use AI to help you write, to guide you through the writing process, rather than letting it write for you, you should be able to continue to develop and keep your own voice as a writer. In an age of increasing use of AI, if everyone uses AI to write, we are likely to all start sounding the same if we just let AI write for us. Wikipedia has even started keeping an updated list of signs of AI writing and researchers to point out that this is already starting to lead to cross-cultural homogenization of not only writing but also the generating of ideas (Agarwal, Naaman, and Vashistha), i.e. to people sounding the same despite different experiences, upbringings, languages, etc. To address this as a writer, you can bring in, for example, your own perspectives and personal experiences to avoid sounding robotic. As a multilingual writer, if you know languages in addition to English, you are encouraged to use them in your writing process, such as for reading sources in those other languages and writing about them in English to deepen your knowledge about the issue being researched. You may specifically prompt the AI to seek sources in other languages that you may know that may offer unique insights that do not exist in English. Valuing and drawing upon your multilingual repertoire to advance your voice with the assistance of AI can enhance your writing, not just by using information from sources in languages other than English, but also by potentially allowing you to communicate your own (and others’) perspectives and lived experiences more accurately, thus strengthening your own voice.

We of course cannot end the chapter without a final note of caution. You can learn more about why some instructors are hesitant to encourage AI use in the writing classroom (see the next chapter in this book); here, we want to emphasize that any engagement with AI in writing warrants some caution. A myriad of costs and negative implications are also associated with AI use, so we don’t want to give the impression that AI is perfect and has no disadvantages, so we encourage you to continue learning and remain critical about AI use. You can also make the decision to entirely avoid its use or to use it sparingly, perhaps in a principled way (whether for writing or in other aspects of your life). According to Joe Wilkins, the CEO of OpenAI, Sam Altman, recently claimed that saying “thank you” and “please” to an AI costs millions of dollars in computing power and associated costs, so we encourage writers to keep this in mind when using AI. Use it when you really need it, in ways (like the ones outlined above) that allow you to maintain agency and control over your writing, in ways that allow you to improve your writing without replacing you as the writer and critical thinker that you are.

AI Use Statement and Author Contributions:

Kamal employed AI in producing a first rough draft of the chapter; specifically, ChatGPT was prompted to generate a chapter outline, including suggested sources. Kamal then reviewed the suggested sources, paraphrased key ideas, added his own original ideas, and wrote a full first draft. Constantin revised the first draft, added his own original ideas and writing, and suggested additional changes. Kamal further revised the chapter manuscript using Constantin’s comments, before final edits by Constantin. Then, editor Daniel Libertz provided feedback on the draft. Another step of revisions followed in response to Daniel’s feedback, led by Constantin with an emphasis on reorganizing and condensing the draft in order to make it more accessible. Kamal then provided his input again. Final copyedits then followed by the Join the Conversation’s editorial team.

End Notes

[1] We use the term AI in this chapter because it is an umbrella term and it is the term that we hear students use the most (if not exclusively) to refer to these technologies, but by and large the systems we refer to are large language models (LLMs) such as ChatGPT. There are of course many other applications/software products used by writers, such as grammar or spell checkers, which also use AI (depending on the product and the definition of AI). Much of the guidance in this chapter applies to such products as well, though the use of some, like the aforementioned grammar/spell checkers, is much more commonly accepted and embraced among writers and instructors of writing.

[2] For a helpful definition see Benavidez (2024).

Works Cited

Armitage, Richard. “Your Brain on ChatGPT.” British Journal of General Practice, vol. 75, no. 758, 2025, p. 410. https://doi.org/10.3399/bjgp25X743181.

Agarwal, Dhruv, et al. “AI Suggestions Homogenize Writing toward Western Styles and Diminish Cultural Nuances.” Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025, pp. 1-21. https://arxiv.org/pdf/2409.11360.

“AI Use Guidance.” Baruch College - Office of the Provost, 2025. https://provost.baruch.cuny.edu/artificial-intelligence-think-tank/ai-use-guidance/.

Allen, Laura K., & Kendeou, Panayiota. “ED-AI Lit: An Interdisciplinary Framework for AI Literacy in Education.” Policy Insights from the Behavioral and Brain Sciences, vol. 11, no. 1, 2024, pp. 3-10. https://doi.org/10.1177/237273222312203.

Benavidez, Rachael. “Socratic Questioning Method.” College Writing I. Course Text for English 110 at CUNY Queens College. Edited by Rachael Benavidez, CUNY, 2024. https://pressbooks.cuny.edu/qcenglish110writingguides/chapter/socratic-questioning-method/.

Cummings, Robert E., et al. “Generative AI in First-Year Writing: An Early Analysis of Affordances, Limitations, and a Framework for the Future.” Computers and Composition, vol. 71, 2024. https://doi.org/10.1016/j.compcom.2024.102827.

Eslami, Maryam, et al. “How Generative AI Is Reshaping Student Writing: A Data-Driven Perspective for Writing Instructors.” Education Sciences, vol. 16, no. 1, 2026. https://doi.org/10.3390/educsci16010001.

Flower, Linda, and John R. Hayes. “A Cognitive Process Theory of Writing.” College Composition and Communication, vol. 32, no. 4, 1981, pp. 365-387. https://doi.org/10.58680/ccc198115885.

Hyland, Ken. “EFL Teachers and Feedback Fatigue: AI to the Rescue?” Language Teaching, vol. 59, no. 3, 2026, pp, 310–326. https://doi.org/10.1017/S0261444825101018.

---. “Writing in the AI Era: Rethinking Writing, Research and Teaching.” Journal of Second Language Writing, vol. 72, 2026. https://doi.org/10.1016/j.jslw.2026.101302.

Jeon, Jaeho, et al. “Generative AI and Its Dilemmas: Exploring AI from a Translanguaging Perspective.” Applied Linguistics, vol. 46, no. 4, 2025, pp. 709–717. https://doi.org/10.1093/applin/amaf049.

Karatay, Yasin., & Leyla Karatay. “Automated Writing Evaluation Use in Second Language Classrooms: A Research Synthesis." System, vol. 123, 2024. https://doi.org/10.1016/j.system.2024.103332.

Kosmyna, Nataliya et al. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv. 31 Dec. 2025. https://doi.org/10.48550/arXiv.2506.08872.[DL5.1]

Marken, Stephanie. “AI Is Routine for College Students, Despite Campus Limits.” Gallup, 2 Apr. 2026. https://news.gallup.com/poll/704090/routine-college-students-despite-campus-limits.aspx.

Meng, Xiao-Li. “Data Science and Engineering with Human in the Loop, behind the Loop, and above the Loop.” Harvard Data Science Review, vol. 5, no. 2, 2023. https://doi.org/10.1162/99608f92.68a012eb.

“National Survey: 95% of College Faculty Fear Student Overreliance on AI and Diminished Critical Thinking among Learners Who Use Generative AI tools.” American Association of Colleges & Universities, 2026. https://www.aacu.org/newsroom/national-survey-95-of-college-faculty-fear-student-overreliance-on-ai-and-diminished-critical-thinking-among-learners-who-use-generative-ai-tools.

Pritts, Nathan. "Student Perspectives on AI-Based Feedback in First-Year Writing: Perceptions and Instructional Implications." International Journal of Innovative Teaching and Learning in Higher Education, vol. 7, no. 1, 2026, pp. 1-19. https://doi.org/10.4018/IJITLHE.413093.

Shaw, Steven. D. & Gideon Nave. “Thinking—Fast, Slow, and Artificial: How AI Is Reshaping Human Reasoning and the Rise of Cognitive Surrender. The Wharton School Research Paper, 2026. https://doi.org/10.31234/osf.io/yk25n_v1.

“Wikipedia: Signs of AI writing.” Wikipedia, The Free Encyclopedia, Wikimedia Foundation. https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing. Accessed 11 Aug. 2026.

Wilkins, Joe. “Sam Altman Admits that Saying "Please" and "Thank You" to ChatGPT Is Wasting Millions of Dollars in Computing Power.” Futurism, Apr. 19 2025. https://futurism.com/altman-please-thanks-chatgpt.

Zhang, Zhe, and Ken Hyland. Student Engagement with Teacher and Automated Feedback on L2 Writing. Assessing Writing, vol. 36, 2018, pp. 90–102. https://doi.org/10.1016/j.asw.2018.02.004.